Published: 23 October 2026
Anthropic said on 22 October that from next week, sessions in the Claude web, desktop and mobile apps will start in auto mode on Pro, Max and Team plans. Claude will no longer pause for your approval at each action. An automated check blocks risky actions instead, and you can switch back to manual at any time.
Enterprise plans keep manual approval as their default. Anthropic has not named a day, so expect the switch in the week of 26 October. The announcement mentions no regional exceptions.
Does auto mode change ordinary Claude chats?
Only where Claude takes actions, such as editing files, using connectors or working in a browser. The approval modes belong to Claude Cowork, the part of Claude that carries out tasks instead of only answering. On Pro and Max, Anthropic is gradually merging chat and Cowork into one Claude, so there the new default covers every conversation in which Claude acts. Team organisations keep chat and Cowork apart for now, so for them it applies to Cowork tasks. A plain question and answer has no step to approve and works as before.
Claude still stops before it deletes files for good
In manual mode, Claude pauses at each action and waits for you to click Allow or Deny. In auto mode it keeps going, while a safety review checks each action first. Anthropic says the review looks for data exfiltration, meaning data sent where it should not go, and for prompt injection: hidden instructions in a web page or file that try to steer Claude.
When the review blocks an action, Claude looks for a safer route or asks you. If it keeps running into blocks, it goes back to asking permission for every step. In every mode, Claude asks before it permanently deletes files.
You change modes with the selector in the message box. A third option, Skip, removes both the questions and the checks. Anthropic advises it only when you fully trust every action, connector, file and app involved.
Anthropic’s test figures come from Claude Code, not from the apps
Claude Code, Anthropic’s coding tool, made auto mode its default on the same plans on 14 August. With that change, Anthropic published a study in which 1,053 paid testers approved commands by hand. Partway through, one prompt was swapped for a clearly dangerous command. The testers caught it 13.6% of the time, while auto mode caught 89%.
Those were developer commands in a test setup. In the apps, the actions are emails, calendar entries, documents and web forms, and Anthropic gives no figures for those. The human score is the real lesson: people who approve dozens of steps a day stop reading them, so a manual default protects less than it seems.
Team admins can keep manual approval with one switch
Team admins find the controls in Organization settings, under Capabilities. “Allow ‘Automatically approve’ mode” is on by default, and turning it off removes auto mode from every member’s selector. “Start new tasks in ‘Automatically approve’ mode” is on for Team plans. Turn it off and new tasks start in manual mode again, while members can still switch within a session.
A separate setting, under Organization settings and then Cowork, covers connector tools that can change data, such as sending an email or updating a record. It is off by default, so members approve those tools once per task. Leave it off until you know which connectors your team uses, because most custom connectors do not mark their tools as read-only. Anthropic’s help page for Team and Enterprise plans lists all three settings.
Keep manual mode for tasks that send, buy or publish
Auto mode suits research, drafting and file work, where a mistake costs you a redo. Anthropic itself advises manual mode when a mistake is hard to undo, such as sending messages or making purchases, or when a task touches sensitive files or accounts. For most organisations that is the line to draw: anything that leaves the building gets a human click.
Does your AI policy say a person approves what an AI tool does before it takes effect? Then auto mode overrides that rule for every Team member next week, unless an admin acts. Turn the auto default off or update the policy, and show staff where the mode selector sits. For uses the AI Act treats as high-risk, such as screening job applicants, the law expects human oversight by trained staff. An automated check is not that, so keep those tasks in manual mode.
For more information, visit the official announcement of the auto mode default on Claude’s account on X.
